There's a middle ground there where failure is an option, but with lots of redundancy and modularity so that such a failure is isolated and quickly resolved (at least temporarily). This is the Erlang/OTP approach to software reliability.